Data to 7 Oct 2026

A Discrete Hard EM Approach for Weakly Supervised Question Answering

arXiv 1909.04849 ↗First posted 11 Sep 2019cs.CL cs.AIcited by 78

Not on the map. Why?

Abstract (arXiv)

Many question answering (QA) tasks only provide weak supervision for how the answer should be computed. For example, TriviaQA answers are entities that can be mentioned multiple times in supporting documents, while DROP answers can be computed by deriving many different equations from numbers in the reference text. In this paper, we show it is possible to convert such tasks into discrete latent variable learning problems with a precomputed, task-specific set of possible "solutions" (e.g. different mentions or equations) that contains one correct option. We then develop a hard EM learning scheme that computes gradients relative to the most likely solution at each update. Despite its simplicity, we show that this approach significantly outperforms previous methods on six QA tasks, including absolute gains of 2--10%, and achieves the state-of-the-art on five of them. Using hard updates instead of maximizing marginal likelihood is key to these results as it encourages the model to find the one correct answer, which we show through detailed qualitative analysis.

Key facts. Cited by 78 papers in the map (data through 7 Oct 2026). First posted on arXiv 11 Sep 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2020: 1; 2021: 31; 2022: 20; 2023: 12; 2024: 8; 2025: 5; 2026: 1.

78 papers.

Sort:

By area:

2020REALM: Retrieval-Augmented Language Model Pre-Trainingcited by 2,126
2023Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflectioncited by 1,800
2021TAPEX: Table Pre-training via Learning a Neural SQL Executorcited by 233
2021PAQ: 65 Million Probably-Asked Questions and What You Can Do With Themcited by 198
2021End-to-End Training of Multi-Document Reader and Retriever for Open-Domain Question Answeringcited by 133
2021Break-It-Fix-It: Unsupervised Learning for Program Repaircited by 71
2021NeurIPS 2020 EfficientQA Competition: Systems, Analyses and Lessons Learnedcited by 52
2022Recent Advances in Text-to-SQL: A Survey of What We Have and What We Expectcited by 47
2021One Question Answering Model for Many Languages with Cross-lingual Dense Passage Retrievalcited by 46
2021UnitedQA: A Hybrid Approach for Open Domain Question Answeringcited by 42
2022A Survey on Table Question Answering: Recent Advancescited by 41
2021Evidentiality-guided Generation for Knowledge-Intensive NLP Taskscited by 40

38 more

2022Learning Math Reasoning from Self-Sampled Correct and Partially-Correct Solutionscited by 38
2023Ever: Mitigating Hallucination in Large Language Models through Real-Time Verification and Rectificationcited by 38
2021R2-D2: A Modular Baseline for Open-Domain Question Answeringcited by 37
2022An Efficient Memory-Augmented Transformer for Knowledge-Intensive NLP Taskscited by 36
2021Capturing Row and Column Semantics in Transformer Based Question Answering over Tablescited by 31
2022ReasTAP: Injecting Table Reasoning Skills During Pre-training via Synthetic Reasoning Examplescited by 29
2021Rider: Reader-Guided Passage Reranking for Open-Domain Question Answeringcited by 28
2021Dual Reader-Parser on Hybrid Textual and Tabular Evidence for Open Domain Question Answeringcited by 25
2021What's in a Name? Answer Equivalence For Open-Domain Question Answeringcited by 21
2022Modeling Multi-hop Question Answering as Single Sequence Predictioncited by 21
2021Weakly-Supervised Hierarchical Models for Predicting Persuasive Strategies in Good-faith Textual Requestscited by 16
2022Grow-and-Clip: Informative-yet-Concise Evidence Distillation for Answer Explanationcited by 16
2022Neural Knowledge Bank for Pretrained Transformerscited by 16
2024Refiner: Restructure Retrieval Content Efficiently to Advance Question-Answering Capabilitiescited by 16
2025RAS: Retrieval-And-Structuring for Knowledge-Intensive LLM Generationcited by 16
2021Narrative Question Answering with Cutting-Edge Open-Domain QA Techniques: A Comprehensive Studycited by 15
2022Chaining Simultaneous Thoughts for Numerical Reasoningcited by 14
2023From Zero to Hero: Examining the Power of Symbolic Tasks in Instruction Tuningcited by 14
2025PhantomWiki: On-Demand Datasets for Reasoning and Retrieval Evaluationcited by 14
2025TempRetriever: Fusion-based Temporal Dense Passage Retrieval for Time-Sensitive Questionscited by 14
2023SUQL: Conversational Search over Structured and Unstructured Data with Large Language Modelscited by 13
2024CtrlA: Adaptive Retrieval-Augmented Generation via Inherent Controlcited by 13
2021Memory and Knowledge Augmented Language Models for Inferring Salience in Long-Form Storiescited by 11
2022Unsupervised Slot Schema Induction for Task-oriented Dialogcited by 11
2021From Good to Best: Two-Stage Training for Cross-lingual Machine Reading Comprehensioncited by 10
2022Can Open-Domain QA Reader Utilize External Knowledge Efficiently like Humans?cited by 10
2024Retrieval is Accurate Generationcited by 10
2022Elaboration-Generating Commonsense Question Answering at Scalecited by 9
2021DoT: An efficient Double Transformer for NLP tasks with tablescited by 8
2021Encoder Adaptation of Dense Passage Retrieval for Open-Domain Question Answeringcited by 8
2023Chain-of-Questions Training with Latent Answers for Robust Multistep Question Answeringcited by 8
2021Learning from Executions for Semantic Parsingcited by 7
2021SPARQLing Database Queries from Intermediate Question Decompositionscited by 7
2021Relation-Guided Pre-Training for Open-Domain Question Answeringcited by 7
2023EM Pre-training for Multi-party Dialogue Response Generationcited by 7
2021Pruning the Index Contents for Memory Efficient Open-Domain QAcited by 6
2022WeCheck: Strong Factual Consistency Checker via Weakly Supervised Learningcited by 6
2023Surpassing GPT-4 Medical Coding with a Two-Stage Approachcited by 6

50 most cited of 78 All 78 (JSON)